THE IMPACT OF ECONOMIC CONDITIONS ON MORTALITY OVER THE LIFETIME
David M. Cutler (Harvard University)
Wei Huang (Harvard University)
Adriana Lleras-Muney (UCLA)
December 2014
Abstract Using historical mortality data covering over 100 birth cohorts in over 20 countries, this paper
examines the contemporaneous and long-term impacts of the economic conditions on mortality. Both
contemporaneous and early-life economic conditions have large but different impact on morality:
contemporaneous recessions are good for health but over the longer term recessions in early life reduce
life expectancy. Although recessions in utero and childhood affect longevity, economic conditions
around the time of graduation (from ages 10 to 30) have the largest impact on lifetime mortality,
particularly for men. We also find strong evidence that the impacts of economic conditions on morality
are much larger and stronger for those countries with lower government expenditure/GDP rates,
suggesting government expenditures can offset the negative effects of economic fluctuations on
outcomes.
1. Data and Methodology
The data we are using are from The Human Mortality Database (HMD), which was created to provide
detailed mortality and population data to researchers, students, journalists, policy analysts, and others
interested in the history of human longevity.1 At present the HMD contains detailed population and
mortality data for 37 countries or areas, including live births, deaths and death rates by sex-age-year cell.
Because of the different availability of the historical data for each country, the starting date for each
country varies. Because we are interested in lifetime effects, we restrict attention to 25 countries with
data in 1950 or earlier. Because old age mortality rates are measured with substantial error, we restrict
the sample to those aged below 90 in this analysis.
1
See http://www.mortality.org/ for more details.
The historical GDP data are from the Gapminder website which combines the data from several
official international statistics and various historical sources.2 That the initial year for the GDP per
capita is 1800 for most of the countries and thus the data cover most of the cohorts we studied since
birth. To de-trend the series, we followed the methodology in van den Berg et al. (2006) by using the
Hodrick-Prescott filter with smoothing parameter 500 to de-trend logarithm of GDP per capita, by
country. Figure 1 shows the case for the United States. The dashed blue line shows the actual logarithm
of GDP per capita over time and the red solid line presents the smoothed one. The de-trended GDP per
capita or GDP fluctuation is defined as the actual value.
We are interested in both the contemporaneous and long-term impacts of the economic conditions
measured by detrended GDP, therefore we kept those cohorts aged over 30. For each birth cohort in
each country, we match the corresponding contemporaneous detrended log GDP per capita as well as
detrended log GDP per capita at age 0 throughout age 30. For simplicity, we aggregate the de-trended
log GDP per capita into 5-year categories: detrended log GDP age 0-5, 6-10 etc.
1.1 OLS models
To estimate contemporaneous and long-term impacts of the economic cycles, we estimate the
following equation:
0−5
6−10
26−30
𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦𝑏𝑐𝑡 = 𝛽0 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑐𝑡 + 𝛽1 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ 𝛽2 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ ⋯ + 𝛽6 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ 𝑋𝑏𝑐𝑡 + 𝜖𝑏𝑐𝑡 (1)
where the dependent variable is the mortality rate for birth cohort b in country c in calendar year t.
𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑐𝑡 is the contemporaneous GDP fluctuation in year t of country c. The coefficients 𝛽0 on it,
capture the impact of contemporaneous GDP fluctuations on mortality. The next few terms in the
𝑥−𝑦
equation, 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
in the equation denotes the mean value of the GDP fluctuations from age x to
age y for birth cohort b in country c. 𝑋𝑏𝑐𝑡 captures a set of control variables including the
contemporaneous smoothed log GDP as well as the fixed effects for calendar year, age, year of birth and
countries. The coefficients on them, 𝛽1 to 𝛽6, capture the long-term impact of the economic conditions
in early life on the mortality of those aged over 30. The observations are weighted by population size.
2
See http://www.gapminder.org/downloads/ for more details.
1.2 Fixed effect models
Using the multi-country year-cohort level data, we are able to control for the country specific economic
conditions for birth cohort when estimating the contemporaneous impacts. Specifically, we can estimate
𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦𝑏𝑐𝑡 = 𝛽0 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑐𝑡 + 𝛿𝑐𝑏 + 𝑋𝑏𝑐𝑡 + 𝜖𝑏𝑐𝑡 (2)
where 𝛿𝑐𝑏 captures the country-birth cohort fixed effects. By doing so, we control for the country
specific contemporaneous economic conditions. By doing so, we actually control for the specific life
experience for the particular birth cohort in each country. The GDP fluctuations in early life are
perfectly collinear with this variable so they are dropped in the main equation.
Similarly, we can also control for the country specific birth cohort fixed effect when estimating the
contemporaneous impact of economic conditions, by estimating the following equation:
0−5
6−10
26−30
𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦𝑏𝑐𝑡 = 𝛽1 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ 𝛽2 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ ⋯ + 𝛽6 ∗ 𝐺𝐷𝑃𝑓𝑙𝑢𝑐𝑏𝑐
+ 𝑋𝑏𝑐𝑡 + 𝛿𝑐𝑡
+ 𝜖𝑏𝑐𝑡 (3)
The difference between equation (2) and equation (1) is the addition of country-calendar year fixed
effects, 𝛿𝑐𝑡 . By including these we control for the country specific contemporaneous economic
conditions—thus in this specification the contemporaneous economic condition are dropped from the
estimation because it is collinear with the newly added dummies.
2. The short term effects of recession
2.1 OLS results
Table 1 presents the results from estimating Equation (1) with all standard errors clustered at country
level. The square term of contemporaneous GDP fluctuation is added to capture possible non-linearities.
First column shows the results for the full sample, and the next two show the results separately by
gender. Consistently, we find the square term on the de-trended log GDP is positive and significant.
Further examination shows that only large deviations (economic booms or large economic recessions)
matter. Between the 5th percentile and 95th percentile of the GDP fluctuation, the relationship between
economic fluctuations and mortality is positive, consistent with the findings in Ruhm (2000).
For the impact of economic conditions in early life, we find consistent negative associations in all three
columns, indicating that the economic booms (recessions) in early life will be negatively (positively)
associated with mortality later. These associations are much more significant for men. Specifically, we
find the most important period in early life is from age 11 to age 20 given the coefficients on the GDP
fluctuations during then are largest in magnitude. The magnitude is large: one point increase in log GDP
at age 11-20 is associated with a 0.6 percentage points decrease in mortality rates after 30.
2.2 FE model for the impact of contemporaneous economic conditions
Table 2 reports the results from the FE model in Equation (2), with country-specific birth-cohort fixed
effects. Panel A show the results for contemporary GDP are consistent with those in Table 1, where we
control for the history of economic conditions since birth instead of using fixed effects. Panel B shows
the results hold if we drop World War I and World War II periods.
Panel C and Panel D show the results by the countries with different level of government expenditures,
measured by the ratio of government expenditure to the total GDP. Because the government
expenditures data are available only for recent years, we just calculate the ratio of government
expenditure to the total GDP from 1998 onwards and use the median level of government expenditure in
the sample as the threshold, which is 0.42 to separate countries into high and low expenditure countries.
The results in Panel C show an insignificant and small impact of contemporaneous GDP fluctuations
among high expenditure countries. But the effects in Panel D (low expenditure countries) are significant
and larger magnitude, suggesting that government expenditure is potentially mitigating the effect of
fluctuations on health.
Furthermore the results in Panels E and F show that the contemporaneous impact of GDP fluctuations is
mostly driven by earlier periods also consistent with the increase in expenditure over time in all
countries. Finally the comparison between Panels G and H indicates that the main effects are driven by
those aged over 60.
3. The impact of economic conditions in early life
Table 3 shows the impact of economic conditions in early life on the mortality of adults ages 30 and
above, controlling for country-calendar year fixed effects. The coefficients, 𝛽1 to 𝛽6, are reported in the
columns with different specifications.
All coefficients are negative, suggesting that unfavorable economic conditions between 0 and 30 lower
subsequent mortality. However the largest effects are for those ages 10-25, these are the years when
individuals enter the labor force. This is particularly true of males for whom all coefficients are negative
and statistically significant. For women the coefficients are smaller, and although the follow the same
age-pattern, only the coefficients from early childhood conditions are statistically significant.
In the next two columns (i.e. column 4 and column 5), we find again that the impact of the early
economic conditions is much larger in countries with lower government expenditure level (relative to
GDP). Differences between columns 4 and column 5 are large and significant, though results are not
reported here. The final two columns, however, we find that the associations are larger for later cohorts,
which is puzzling. Analysis in the future will shed some light on these issues and phenomenon.
Table 1: Impact of Contemp. And Earlier life GDP fluc. On Mortality rates aged over 30
(1)
(2)
(3)
Dependent var. is Mortality rates in percentage
VARIABLES
All population
Men
Women
Contemporaneous GDP fluc.
GDP Fluc.
GDP Fluc. Square
GDP Fluc.at ealier life
Age 0
Age 1 - 5
Age 6 - 10
Age 11 - 15
Age 16 - 20
Age 21 - 25
Age 26 - 30
Observations
R-squared
Mean of Y
SD of Y
0.0582
(0.113)
1.673**
(0.799)
-0.100
(0.115)
1.846*
(0.998)
0.179
(0.128)
1.607**
(0.706)
-0.125**
(0.0446)
-0.247**
(0.100)
-0.329*
(0.161)
-0.406*
(0.206)
-0.393*
(0.227)
-0.267
(0.158)
-0.0767
(0.155)
-0.144***
(0.0448)
-0.327***
(0.106)
-0.466***
(0.156)
-0.608***
(0.203)
-0.638***
(0.206)
-0.490***
(0.139)
-0.304**
(0.146)
-0.105**
(0.0429)
-0.177*
(0.0986)
-0.221
(0.171)
-0.245
(0.220)
-0.170
(0.245)
-0.112
(0.180)
0.0644
(0.185)
132,288
0.955
1.570
2.582
132,288
0.964
1.765
2.792
132,288
0.945
1.393
2.488
Contemp. GDP level
Yes
Yes
Smoothed GDP before 30
Yes
Yes
Calendar Year FE
Yes
Yes
Age in years FE
Yes
Yes
Year of Birth FE
Yes
Yes
Country FE
Yes
Yes
Notes: Robust standard errors in parentheses are clustered at country level.
*** p<0.01, ** p<0.05, * p<0.1
Yes
Yes
Yes
Yes
Yes
Yes
Table 2 Contemporaneous Economic Conditions and Death Rates (FE model)
(1)
(2)
(3)
(4)
(5)
(6)
Dependent var. is Mortality rates in percentage
VARIABLES
All population
Men
Women
All population
Men
Women
Panel A. Full sample
Panel B. Drop WW1 and WW2 period
Contemporaneous GDP fluc.
GDP Fluc.
0.137
0.0146
0.229
0.334
0.307
0.296
(0.177)
(0.190)
(0.191)
(0.268)
(0.224)
(0.307)
GDP Fluc. Square
1.706**
1.509
1.963**
2.282**
2.392*
2.330*
(0.737)
(0.927)
(0.707)
(1.071)
(1.226)
(1.172)
Observations
132,288
132,288
132,288
Panel C. Higher Government Expenditure
Countries
Contemporaneous GDP fluc.
GDP Fluc.
GDP Fluc. Square
Observations
Contemporaneous GDP fluc.
GDP Fluc.
GDP Fluc. Square
Observations
Contemporaneous GDP fluc.
GDP Fluc.
GDP Fluc. Square
Observations
0.120
(0.197)
1.512
(1.195)
0.0122
(0.217)
1.232
(1.247)
0.207
(0.208)
1.607
(1.204)
67,660
67,660
67,660
Panel E. Earlier Years (Years before meidan in
each specific country)
0.425
(0.337)
2.286**
(0.996)
93,313
0.425
(0.388)
2.560**
(1.101)
122,808
122,808
122,808
Panel D. Lower Government Expenditure
Countries
0.479*
(0.241)
1.716**
(0.609)
0.242
(0.184)
1.427**
(0.646)
0.661*
(0.309)
1.956**
(0.719)
61,124
61,124
61,124
Panel F. Later Years (Years after meidan in
each specific country)
0.431
(0.308)
2.029**
(0.904)
-0.111
(0.191)
0.950
(0.809)
-0.158
(0.208)
0.624
(0.937)
-0.0692
(0.212)
1.150
(0.818)
93,313
93,313
Panel G. Aged 30 - 60
92,145
92,145
Panel H. Aged 61+
92,145
-0.0628
(0.0778)
0.577
(0.460)
-0.188
(0.139)
0.550
(0.604)
0.0439
(0.0336)
0.620
(0.369)
-0.314
(0.571)
5.206***
(1.636)
-0.265
(0.592)
5.427***
(1.908)
-0.228
(0.615)
5.133***
(1.620)
68,899
68,899
68,899
63,389
63,389
63,389
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Covariates controlled for
Comtemp. smoothed GDP
Yes
Yes
Yes
Calendar Year FE
Yes
Yes
Yes
Age FE
Yes
Yes
Yes
Year of Birth FE
Yes
Yes
Yes
Country FE
Yes
Yes
Yes
Country*Birth year FE
Yes
Yes
Yes
Notes: Robust standard errors in parentheses are clustered at Country level.
*** p<0.01, ** p<0.05, * p<0.1
Table 3 Impact of Economic Conditions on mortality at later ages (30+)
(1)
(2)
(3)
(4)
(5)
Dependent variable is Death Rates in Percentage
Sample
Full sample
High Gov. exp Ctys Low Gov. exp Ctys
VARIABLES
All
Men
Women
All
All
GDP Fluc. at birth
GDP Fluc. 1-5
GDP Fluc. 6-10
GDP Fluc. 11-15
GDP Fluc. 16-20
GDP Fluc. 21-25
GDP Fluc. 26-30
Observations
R-squared
Mean of Y
SD of Y
(6)
(7)
Ealier Cohorts
All
Later Cohorts
All
-0.129**
(0.0616)
-0.279*
(0.145)
-0.414*
(0.223)
-0.517*
(0.285)
-0.518
(0.302)
-0.363*
(0.207)
-0.242
(0.217)
-0.119*
(0.0628)
-0.298*
(0.150)
-0.458**
(0.211)
-0.613**
(0.274)
-0.661**
(0.275)
-0.502**
(0.187)
-0.432*
(0.209)
-0.137**
(0.0563)
-0.267*
(0.140)
-0.396
(0.234)
-0.456
(0.297)
-0.390
(0.320)
-0.288
(0.230)
-0.138
(0.246)
-0.108***
(0.0252)
-0.250***
(0.0605)
-0.257*
(0.134)
-0.289
(0.167)
-0.176
(0.191)
-0.0961
(0.174)
0.0505
(0.182)
-0.223**
(0.0935)
-0.484**
(0.171)
-0.613**
(0.208)
-0.833**
(0.303)
-0.824**
(0.267)
-0.567**
(0.252)
-0.617
(0.373)
-0.0345
(0.0941)
-0.198
(0.171)
-0.127
(0.212)
-0.212
(0.236)
-0.151
(0.202)
-0.113
(0.156)
-0.0891
(0.109)
-0.348***
(0.0885)
-0.607***
(0.159)
-1.042***
(0.315)
-1.279***
(0.395)
-1.262***
(0.386)
-0.954***
(0.322)
-0.493
(0.377)
132,288
0.958
1.570
2.582
132,288
0.967
1.765
2.792
132,288
0.948
1.393
2.488
67,660
0.959
1.762
2.811
61,124
0.969
1.419
2.370
66,732
0.972
1.819
2.779
65,556
0.977
1.414
2.438
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Covariates controlled for
Age FE
Yes
Yes
Yes
Birth cohort FE
Yes
Yes
Yes
Country FE
Yes
Yes
Yes
Year FE
Yes
Yes
Yes
Country-Year FE
Yes
Yes
Yes
Notes: Robust standard errors in parentheses are clustered at Country level.
*** p<0.01, ** p<0.05, * p<0.1
8
9
10
11
United States
7
GDP per capita
Smoothed value
1800
1850
1900
year
1950
2000
Figure 1 Logarithm of GDP per capita and the Smoothed Value, the United States 1800-2012
Notes: GDP Data comes from Gapminder (http://www.gapminder.org/downloads/). The smoothed GDP is derived from
Hodrick-Prescott filter with smoothing parameter 500. The de-trended GDP is defined as the actual value (Blue line)
subtracting the smoothed value (Red line).